PDXK mutations cause polyneuropathy responsive to pyridoxal 5′‐phosphate supplementation

Objective To identify disease‐causing variants in autosomal recessive axonal polyneuropathy with optic atrophy and provide targeted replacement therapy. Methods We performed genome‐wide sequencing, homozygosity mapping, and segregation analysis for novel disease‐causing gene discovery. We used circular dichroism to show secondary structure changes and isothermal titration calorimetry to investigate the impact of variants on adenosine triphosphate (ATP) binding. Pathogenicity was further supported by enzymatic assays and mass spectroscopy on recombinant protein, patient‐derived fibroblasts, plasma, and erythrocytes. Response to supplementation was measured with clinical validated rating scales, electrophysiology, and biochemical quantification. Results We identified biallelic mutations in PDXK in 5 individuals from 2 unrelated families with primary axonal polyneuropathy and optic atrophy. The natural history of this disorder suggests that untreated, affected individuals become wheelchair‐bound and blind. We identified conformational rearrangement in the mutant enzyme around the ATP‐binding pocket. Low PDXK ATP binding resulted in decreased erythrocyte PDXK activity and low pyridoxal 5′‐phosphate (PLP) concentrations. We rescued the clinical and biochemical profile with PLP supplementation in 1 family, improvement in power, pain, and fatigue contributing to patients regaining their ability to walk independently during the first year of PLP normalization. Interpretation We show that mutations in PDXK cause autosomal recessive axonal peripheral polyneuropathy leading to disease via reduced PDXK enzymatic activity and low PLP. We show that the biochemical profile can be rescued with PLP supplementation associated with clinical improvement. As B6 is a cofactor in diverse essential biological pathways, our findings may have direct implications for neuropathies of unknown etiology characterized by reduced PLP levels. ANN NEUROL 2019;86:225–240

P eripheral neuropathies are among the most common neurological disorders worldwide, affecting approximately 20 million people in Europe and the USA alone, with few disease-modifying treatments established for these conditions. [1][2][3] Inherited forms of peripheral neuropathies affecting both the motor and sensory nerves, known as Charcot-Marie-Tooth disease (CMT), are the most common genetic neuromuscular disorders, affecting approximately 1 in 2,500 people. [4][5][6][7][8] However, only 25% of all autosomal recessive (AR) CMT cases have causal variants identified, 9 and treatment is only supportive. [1][2][3] Most known genes associated with AR CMT cause disease either by affecting the cytoskeleton or axonal trafficking. 2

Study Participants
Patients with polyneuropathy and optic atrophy were identified by neurogeneticists at the National Hospital for Neurology and Neurosurgery London and Children's Hospital of Eastern Ontario. The affected cases were recruited along with unaffected family members under institutional review board/ethics-approved research protocols (University College London Hospitals: 04/N034; Children's Hospital of Eastern Ontario: CTO/1577) with informed consent. All cases had extensive genetic, metabolic, and mitochondrial investigations carried out that excluded acquired and other inherited causes of polyneuropathy and optic atrophy.

Phenotype and Clinical Measures
All cases had comprehensive phenotyping performed by neurogenetics specialists including clinical assessment, electrophysiology, neuroimaging, and videography. Motor function was quantified using the Medical Research Council (MRC) muscle scale, a standardized scale for the assessment of the peripheral nervous system (PNS) that grades muscle power from 0 to 5 in 6 validated movements and provides a total score of a possible 60 points. The disease severity and progression were assessed using the validated Neurological Impairment Scale (NIS) 10 and the Charcot-Marie-Tooth Neuropathy Score version 2 (CMTNS2). CMTNS2 is a reliable and valid composite of symptoms (3 items), signs (4 items), and neurophysiology (2 items). It is designed to measure length-dependent motor and sensory impairment in genetic neuropathies. 11,12 The NIS forms part of the standard minimum dataset for the Functional Independence Measure and UK Functional Assessment Measure. It records the severity of functional impairment (rated 0-3) across 13 domains mapped onto the International Classification of Functioning. The score range is 0 to 50. The affected individuals from Family 1 had MRC scales, CMTNS2, and NIS assessment before and after replacement therapy.

Genome-wide Sequencing and Haplotype Analysis
DNA was extracted from peripheral blood. Whole genome sequencing (WGS) was performed by deCODE genetics (Reykjavík, Iceland), using DNA from the 2 affected siblings from Family 1. A HiSeq4000 instrument (Illumina, San Diego, CA) was used to generate 100bp paired-end reads. Alignment was performed using BWA (http://bio-bwa.sourceforge.net/) 13 with GRCH37 as a reference. Variants were called using the GATK [14][15][16][17] Unified Genotyper-based pipeline [14][15][16] workflow. All variants were annotated using ANNOVAR 18 and filtered using custom R scripts. Shared regions of homozygosity were identified using HomozygosityMapper 19 and Bcftools/RoH. 20 Only novel or very rare variants with a minor allele frequency of <0.01 in the 1000 Genomes Project, 21 NHLBI GO Exome Sequencing, 22 and Exome Aggregation Consortium database 23 226 Volume 86, No. 2 were included. Coding/splicing homozygous variants that were within the autozygome of the affected cases were prioritized. Whole exome sequencing (WES) was performed in Family 2 using DNA from both affected siblings and their unaffected mother as part of the Care4Rare Canada Consortium. Exonic DNA was selected using the SureSelect Clinical Research Exome V1 kit (Agilent Technologies, Santa Clara, CA), then sequenced on an Illumina NextSeq 500 with 2 × 150bp chemistry. Read alignment, variant calling, and annotation were done as outlined for previous FORGE and Care4Rare Canada projects. 24 Average coverage for the exomes was 182×, 190×, and 244×, and >97% of consensus coding sequence exons in all exomes were covered at >10×.

Sanger Sequencing
The variants identified by WGS and WES were confirmed by Sanger sequencing. The region was amplified using the following primers: 5 0 -AGGAGGATCAGGGATGGGAG-3 0 and 5 0 -CTCTCATAT CCTGCTCCCCA-3 0 in Family 1 and 5 0 -CTTGCCGCTCTAT GTTTGGAC-3 0 and 5 0 -GAGAGTGACAGGCGCAACTC-3 0 in Family 2. The promoter region of PDXK was amplified using the following primers: 5 0 -GCGGTTCCCTTGGGTATC-3 0 and 5 0 -ACG CCTCCTTCTGACCTC-3 0 . Genomic DNA was extracted from dried blood spots (DBSs) from affected individuals from Family 1 and controls using the QIAamp DNA Micro Kit (Qiagen, Valencia, CA) and from blood from Family 2. Sequencing reactions were performed using the BigDye Terminator 1.1 system (Thermo Fisher Scientific, Paisley, UK) followed by sequencing using an ABI DNA Analyser (Thermo Fisher Scientific). Electropherograms were analyzed using the Sequencher software package (Gene Codes Corporation, Ann Arbor, MI).

Gene Co-expression Analysis
We generated gene co-expression networks for central nervous system (CNS) and PNS tissue-specific transcriptomic data generated by the Genotype-Tissue Expression Consortium (GTEx7 V6 gene expression data are accessible from https://www. gtexportal.org/). 25 We analyzed each of the tissue sample sets separately by filtering genes on the basis of an Reads Per Kilobase Million (RPKM) >0.1 (observed in >80% of the samples for a given tissue). We then corrected for batch effects, age, gender, and RNA integrity numbers (RIN) using ComBat. 26 Finally, we used the residuals of these linear regression models to construct gene co-expression networks for each tissue using WGCNA 27 and postprocessing with k-means 28 to improve gene clustering. Gene modules were functionally annotated with gProfileR4 R package using the Gene Ontology (GO) database without electronic inferred annotations and accounting for multiple testing with gSCS. 29 The top-down plot figure was done with Cytoscape 3.5.1. 30 We initially selected the 50 most connected genes within the black module, for the coexpression network of tibial nerve, based on adjacency values. We included all connections between genes. Then we filtered out the connections with weight <0.05. Finally, we applied a Kamada-Kawai layout algorithm to the remaining genes and edges to dispose them spatially at the canvas so their relative positions reflected similarity in connections.

Plasmid Constructs
The human PDXK wild-type (WT) and PDXK mutant p.Ala228Thr cDNA were purchased from Invitrogen (Carlsbad, CA) and cloned into the plasmid pGEX6p-1 (GE Healthcare, Piscataway, NJ) using the sites BamHI and NotI (New England Biolabs, Ipswich, MA).
Ultraviolet (UV) and CD spectra were acquired on the Applied Photophysics (Leatherhead, UK) Chirascan Plus spectrometer across the 400 to 190nm wavelength region. Tenmillimeter (400-230nm) and 0.5mm (260-190nm) strain-free rectangular cells were employed. The instrument was flushed continuously with pure evaporated nitrogen throughout the experiment. The following parameters were employed: 2nm spectral bandwidth, 1nm step size, and 1-second instrument measurement time-per-point. All spectra were acquired at 25 C, and buffer baseline corrected. Light scattering correction was August 2019 227 applied on the UV spectra of both proteins using the Chirascan Pro-Data Software (Applied Photophysics). Where possible, spectra were smoothed with a window factor of 4 using the Savitzky-Golay method 32 for better presentation. The far-UV CD spectra of the 2 proteins were corrected for concentration and pathlength and expressed in terms of Δε (M −1 cm −1 ) per amino acid residue. Protein secondary structure content was assessed using the Principle Component Regression method based on 16 known protein structures 31 embedded in the PLSPlus/IQ routine in GRAMS32 AI software (Galactic Software, Salem, MA).

Isothermal Titration Calorimetry
The affinity of a nonhydrolyzable analogue of adenosine triphosphate (ATP), called adenosine 5 0 -(3-thiotriphosphate) tetralithium salt (ATPγS), for both PDXK WT and p.Ala228Thr mutant, was examined using a MicroCal Isothermal Titration Calorimeter ITC200 instrument (Malvern Panalytical, Malvern, UK). Both proteins were prepared as described above in 30mM tris, pH 7, 100mM KCl, 3mM MgCl 2 , 0.5mM TCEP (isothermal titration calorimetry [ITC] buffer). An initial 0.2M stock of ATPγS was prepared in water and diluted with ITC buffer to a final concentration of 250μM. The concentration was adjusted using the UV absorbance at 259nm, and the extinction coefficient of 15.4mM −1 cm −1 suggested by the ligand provider (Jena Bioscience, Jena, Germany).
The experiments were conducted at 25 C following standard procedures as described previously. 33 Forty microliters of 250μM ATPγS was titrated into a 330μl solution of 25μM WT or p.Ala228Thr protein. The ITC experiment consisted of twenty 2μl injections with a spacing of 180 seconds. Heat produced by titrant dilution was obtained by a control experiment, titrating into buffer alone, under the same conditions. The MicroCal-Origin 7.0 software package was used to fit the integrated heat data obtained for the titrations corrected for heats of dilution, using a nonlinear least-squares minimization algorithm based on an independent binding sites model. ΔH (reaction enthalpy change in kcal/mol), Ka (1/K D ; equilibrium association constant per molar), and n (molar ratio of the proteins in the complex) were the fitting parameters. The reaction entropy was calculated using the relationships ΔG = ΔH − TΔS = −RTln(1/K D ).

Western Blotting
Fibroblast samples from affected, carriers, and healthy people were obtained with consent. After reaching 90% confluence, the cultured cells were harvested and lysed in ice-cold lysis radioimmunoprecipitation assay buffer containing 50mM tris-HCl pH 8.0, 150mM NaCl, 0.5% DOC, 0.1% Triton X-100, 0.1% sodium dodecyl sulphate (SDS), and SIGMAFAST Protease Inhibitor Cocktail for 30 minutes. The lysate was then centrifuged at 16,000 × g at 4 C, and the supernatant transferred to a new tube. The protein concentration was measured using the BCA Protein Assay Kit (Thermo Fisher Scientific) according to manufacturer's protocol. Equal amounts of protein (30μg) from affected and healthy individuals were separated on a 10% SDS-polyacrylamide gel electrophoresis gel (NuPAGE, Invitrogen) and transferred onto an Immobilon membrane (Millipore, Billerica, MA). After blocking the membrane with 2% fat-free milk in phosphate-buffered saline-Tween for 1 hour at RT, the membrane was incubated with anti-PDXK antibody (Abcam, Cambridge, UK; 38208, 1:500) for 1 hour at RT. The membranes were subsequently washed and incubated with horseradish peroxidase-conjugated goat antirabbit IgG (Bio-Rad Laboratories, Hercules, CA; 17210, 1:5,000) for 1 hour at RT. Blots were developed using ECL Prime (GE Healthcare), visualized via a ChemiDoc Touch Imaging System (Bio-Rad Laboratories), and analyzed using Image Lab 5.2 software (Bio-Rad Laboratories). For the quantification, the signal intensity of each band corresponding to the PDXK WT and PDXK p.Ala228Thr mutant protein was normalized to the signal intensity of the corresponding reversible Ponceau staining as a reliable loading control. 34 The amount of PDXK WT and mutant protein was expressed as a percentage of the control sample. The experiment was repeated 3 times.

Measurement of Pyridoxal Kinase Activity
The pyridoxal (PL) kinase activity present in DBSs was determined by adapting the protocol used by Wilson et al, 35 with the following modifications. Three-millimeter discs punched from DBSs were incubated for 10 minutes at 37 C with shaking at 300rpm in a reaction buffer containing 20mmol/l potassium phosphate adjusted to pH 6.1, 10μmol/l PL, and 300μmol/l MgATP (all purchased from Sigma-Aldrich). The same method was adapted for the investigation of activity of the purified recombinant PL kinase. Five microliters of 20ng/μl purified PL kinase protein lysate was placed in an ultraperformance liquid chromatography (UPLC) 96-well sample plate (Waters, Elstree, UK). Subsequently, 115μl of reaction buffer containing potassium phosphate (pH 6.1), PL, and MgATP was added to create a final concentration of 20mmol/l potassium phosphate, with PL and MgATP concentrations adjusted as appropriate. Investigation of PL kinase kinetics as a function of PL concentration used an MgATP concentration of 300μmol/l. A PL concentration of 50μmol/l was utilized for the determination of PL kinase kinetics as a function of MgATP concentration. After reaction buffer addition, the plate was incubated for 10 minutes at 37 C in an Eppendorf (Hamburg, Germany) Thermomixer C with shaking at 300rpm prior to addition of 120μl of a reaction stop mix identical to that used for the determination of pyridoxamine (PM) 5 0 -phosphate (PMP) oxidase activity from DBSs.
DBSs were collected from 4 affected cases and 22 controls (age = 15-92 years). PL kinase activity of the recombinant proteins and that of DBSs were determined by using UPLC-tandem mass spectrometry (MS/MS) to determine the formation of PL 5 0 -phosphate (PLP) after incubation of these with the enzyme substrate PL and expressed as μmol PLP l −1 h −1 from recombinant protein and as pmol PLP (3mm DBS) −1 h −1 from DBSs.
Data collection and statistical analysis were performed using Waters MassLynx and GraphPad (San Diego, CA) Prism 6.0 software packages. 228 Volume 86, No. 2 UPLC-MS/MS Measurement of the B 6 Vitamers B 6 vitamer concentrations (pyridoxine, PL, PLP, PM, PMP) and pyridoxic acid were quantified by UPLC-MS/MS using stable isotope-labeled internal standards. 35,36 High-Performance Liquid Chromatograph Measurement of B 6 Vitamers PL phosphate was measured by high-performance liquid chromatography (HPLC) with fluorescence detection using a Chromsystems Instruments & Chemicals (Gräfelfing, Germany) kit as described previously. 35 Pyridoxic acid concentration was calculated using an external calibration standard.

PLP Measurements before and after PLP Replacement
Plasma PLP concentrations before and after PLP replacement were measured using 2 independent HPLC and UPLC-MS/MS methods as described above to validate the results.
The difference between groups was tested with the use of a 1-way analysis of variance test, followed by Tukey-Kramer test.

Measurement of Neurofilament Light Chain Concentration
Plasma neurofilament light chain (NFL) concentration was measured by digital enzyme-linked immunosorbent assay using the commercially available NF-Light kit on a single molecule array (Simoa) platform (HD-1 Analyzer) according to instructions by the manufacturer (Quanterix, Lexington, MA). All measurements were performed in duplicates, in 1 round of experiments using 1 batch of reagents. Intra-assay coefficients of variation were <10%.

Disease Characterization
Family 1 is of Cypriot origin (affected individuals F1-II-5, currently 80 years old, F1-II-6, currently 75 years old, and F1-II-9, died at 71 years old from an unrelated medical condition). Three affected siblings presented with difficulties running, distal wasting, and weakness in the lower limbs before the age of 10 years, progression to upper limbs by the age of 12 years, and onset of visual loss in their 40s. The disease progressed, and at the age of 79 years, case F1-II-5 required the assistance of 2 people for sit-to-stand, mobilizing with the assistance of 2 plus Zimmer frames for transfers and short distances, and using a wheelchair for all other activities. Neurogenic-type pain, particularly in the lower limbs, was difficult to control despite multiple lines of symptomatic treatment. Case F1-II-6 (age = 74 years) could only walk short distances with assistance from 1 individual. Neurological examination showed bilateral distal wasting from mid forearm and below the knee, with wasting of the thenar, hypothenar, and intrinsic muscle of the hands and distal lower limb muscles, pes cavus, and absent reflexes throughout in both patients (see Fig 1B). The MRC muscle score was 44 (of 60) in cases F1-II-5 and F1-II-6. Strikingly, both patients had severely reduced sensation in all limbs and in all modalities, with profound joint position sense loss up to the knee and vibration loss up to the iliac crest in both cases (see Table 1 and Supplementary Table 1). Vision was reduced (including color vision) and fundoscopy confirmed bilateral optic atrophy in both cases (see Fig 1C). The remaining cranial nerve examination was normal. The third sibling F1-II-9 (now deceased) was similarly affected.
Family 2 is of Scottish (paternal) and Italian (maternal) origin with 2 affected sisters (Patient F2-II-1, currently 31 years old; Patient F2-II-2, currently 29 years old) who presented with childhood onset of a mild decrease in balance, which progressed to difficulty with running in adolescence, and then to mild distal hand weakness in the 3rd decade. Both sisters ambulate independently without aids, and neither sister has symptoms of visual loss at their present ages. Neurological examination showed mild symmetrical distal wasting in hand intrinsic muscles and prominent distal lower limb muscle wasting with pes cavus and mild hammertoes. There was mild distal weakness in hand intrinsic muscles, in the upper limbs, and in the toes and distal lower legs and feet. The MRC scores were 56 (F2-II-1) and 54 (F2-II-2). Both sisters were areflexic, with reduced pinprick sense in the lower extremities and decreased vibration sense to the knees (see Table 1 and Supplementary Table 1). Color vision was reduced. Fundoscopy confirmed bilateral optic atrophy in both cases; the remainder of the cranial nerve examination was normal.
Nerve conduction studies (NCS) in all affected cases revealed a progressive, length-dependent, sensorimotor, predominantly axonal neuropathy (Supplementary Table 2). Early in the disease (as seen in Family 2), the NCS show a predominantly axonal neuropathy affecting mainly the lower limbs with absent sensory responses. Mild conduction slowing with normal motor responses was recorded in the upper limbs. The natural history of this disorder in the older individuals from Family 1 (F1-II-5 and F1-II-6) with data recorded over 20 years of disease duration suggests that untreated, the neuropathy progresses in a length-dependent fashion and is characterized by prolonged distal motor latency (DML), severe reduction of compound muscle action potentials (CMAPs), and absent sensory responses affecting upper and lower limbs, as confirmed on NCS. Interestingly, both patients present with significantly more prolonged DML in the median compared to ulnar nerve August 2019 229 (6.1 vs 3.6 milliseconds). Although it is possible that this is secondary to a longstanding median compression lesion at the wrist, we cannot exclude that this finding is diseaserelated. The reduced CMAPs were associated with reduction of conduction velocities in the intermediate range mainly in the median nerves. Electromyography in all cases tested was consistent with chronic denervation with no myopathic changes.
To further ascertain the exact type of neuropathy, we performed a nerve biopsy in case F1-II-6. It showed diffuse and severe depletion of both small and large myelinated axons with regenerating clusters but no "onion . PLP re-enters the circulation bound to a lysine residue of albumin. Homeostatic regulation of tissue levels of PLP is achieved by various mechanisms, including feedback inhibition of PNPO and PL kinase by PLP. 62 Albumin in plasma, hemoglobin in erythrocytes, and glycogen phosphorylase in muscle also play a role, binding to PLP and helping to keep concentrations of this very reactive aldehyde low 63 so as to avoid any unwanted reactions with biologically important molecules. Subsequent delivery of PLP to the tissues requires hydrolysis of circulating PLP to PL by the ectoenzyme tissue nonspecific alkaline phosphatase (TNSALP). The resulting pyridoxal is able to enter cells prior to being rephosphorylated by PL kinase to produce the active cofactor, PLP, required by B 6 -dependent apoenzymes. 64 Within cells, recycling pathways also exist, with PMP being oxidized by PNPO to form PLP. 63   bulbs." This confirms a longstanding axonopathy with no overt demyelination (see Fig 1D). All cases reported here had pale optic discs on clinical examination. We performed visual evoked potentials (VEPs) to further describe the involvement of the optic nerve in the course of disease. We found that VEPs in Case F1-II-6 had decreased considerably in amplitude from 7.9 and 7.4 in each eye to 1.8 and 1.9 over a 20-year period, indicating axonal loss and bilateral visual pathway involvement (see Supplementary Table 2). Similar results were confirmed in case F1-II-5. Family 2 has not been investigated with VEPs.

Identification of Disease-Causing PDXK Variants
Previous extensive genetic, 37 metabolic, and mitochondrial investigations failed to identify any acquired or known causes of inherited polyneuropathy in all cases. WGS and homozygosity mapping in Family 1 revealed a shared homozygous region from 42 to 45Mb on chromosome 21. This region included a shared homozygous variant in the 2 affected siblings that was carried by unaffected family members in heterozygous state: PDXK (NM_003681) c.682G>A (p.Ala228Thr). This variant was absent in 150 Cypriot controls. The p.Ala228Thr is absent from the-gnomAD database in homozygous state though reported in 7 heterozygous of atotal of 250586 alleles (total allele frequency of 0.00002793). 23 Independently, homozygosity mapping in Family 2 revealed a 1.35Mb region of homozygosity on chromosome 21. WES in Family 2 identified a shared homozygous variant in this region, the c.659G>A p.(Arg220Gln) in the 2 affected sisters, carried by each parent. This variant is absent from gnomAD in homozygous state and reported in 5 heterozygous European alleles from 250146 (total allele frequency of 0.00001999). 23 It is predicted to be deleterious to protein function by in silico programs (Combined Annotation Dependent Depletion score = 24). Sanger sequencing in affected individuals and carriers confirmed both variants (see Fig 1E).
PDXK is ubiquitously expressed in humans encoding for PL kinase (PDXK), a cytoplasmic protein that converts vitamin B 6 to its active form, PLP (see Fig 1F). PLP is an essential cofactor for >70 human enzymes representing diverse, essential biological pathways including amino acid and neurotransmitter metabolism.
Because PDXK mutations produce a peripheral rather than a predominantly CNS disease, we investigated these differences further in silico using public bulk tissue 25 and cell-specific transcriptomic data. 38 This demonstrated that human peripheral (tibial-mixed motor and sensory) nerve is among the tissues with the highest expression of PDXK (Fig 2A), with evidence for expression specifically in peripheral sensory neurons, based on single cell mouse transcriptomic data (see Fig 2B). 38 This showed different co-expression profiles in brain compared to peripheral nerve tissue. Weighted gene co-expression analysis demonstrated that in 9 of 11 brain regions analyzed, PDXK was coexpressed with genes relating to synaptic transmission and neuronal identity, whereas co-expression data from the tibial nerve showed that PDXK was located within a module enriched for genes involved in oxidation-reduction processes (GO: 055114; false discovery rate-corrected p = 2.36 × 10 −9 ; see Fig 2C and Supplementary Table 3).

Assessment of the Impact of PDXK Variants on Protein Structure and Enzymatic Activity
Arginine 220 and alanine 228 are evolutionarily conserved amino acids in eutherians (see Fig 2D). Arginine 220 is located in the β9 in the vicinity of the ATP-binding site. 39 Alanine 228 is located in the ATP-binding pocket of PL kinase (see Fig 2E), a critical region of the kinase, part of the loop region between strand β11 and helix α7 that rotates away from the active site, allowing binding of the ATP adenine. 39 Although the Ala228 side-chain does not contact ATP directly, its backbone is involved in recruiting ATP to the catalytic pocket.
Far-UV CD spectra analysis of both WT and p.Ala228Thr revealed that, whereas both proteins were folded, there were small but significant differences in their secondary structure (Fig 3A). WT PDXK had an estimated α-helix and β-strand content of 31.2% and 19%, respectively, similar to previous X-ray studies (PDB accession number 3KEU). 40 In the mutant protein, 26.0% and 21.1% of helical and strand content, respectively, were estimated, suggesting altered protein conformation. Furthermore, near-UV CD analysis revealed distinctive CD fingerprints for WT and p.Ala228Thr proteins, indicating conformational rearrangements of aromatic side-chains 41 around the catalytic pocket, most likely hindering the enzyme's ability to bind ATP.
We performed ITC confirming that p.Ala228Thr affected the ATP-binding ability of PDXK. Interaction of WT PDXK with the nonhydrolyzable ATP analogue ATPγS generated well-interpolated, sigmoid-shaped curves, based on an independent and equivalent binding site model centered on 1:1 stoichiometry. By contrast, no association between p. Ala228Thr and ATPγS was observed (see Fig 3B).
Whereas no reduction in the expression of the mutant protein (p.Ala228Thr) compared to WT PDXK in case-derived fibroblasts was evident (see Fig 3C), kinetic studies with human recombinant PDXK showed an approximate 2-fold increase in the K m of the p.Ala228Thr mutant protein (31.9μmol/l) for PL compared to WT protein (14.5μmol/l) and a reduction in the V max of the mutant protein (p.Ala228Thr) to 0.95μmol/l/h compared to 2.52μmol/l/h for WT PDXK. Upon variation of MgATP concentration, sigmoidal kinetics were observed (WT: k 0.5 = 53.4μmol/l, V max = 16.8pmol/h; p.Ala228Thr: k 0.5 = 174.4μmol/l, V max = 6.3pmol/h), indicating cooperative binding between the 2 substrates, ATP and PL (see Fig 3D, E).
Analysis of erythrocyte PDKX activity from cases carrying the p.Arg220Gln and p.Ala228Thr substitutions further confirmed the deleterious effect of these variants on PL kinase activity (see Fig 3F). The PDKX activity in the 4 affected cases was reduced (1.1 and 0.8pmol DBS −1 h −1 in Family 1 and 1.19 and 0.5pmol DBS −1 h −1 in Family 2) relative to controls (2.6-14.7pmol DBS −1 h −1 , mean = 8.0). Only the most connected genes are shown. PDXK gene is highlighted in yellow. Genes known to be associated with the Gene Ontology term GO:0055114, oxidation-reduction process, are highlighted in red. The size of gene nodes reflects their connectivity with the rest of the genes in the module. PDXK is among the top 60 most connected genes. Proximity of genes in the plot reflects their similarity in terms of shared connections with other genes. Interestingly, within the PDXK regulon from the tibial nerve, we found DHTKD1 already linked to Mendelian disorders and associated with primary peripheral axonal neuropathy. 59,60 (D) Conservation of p. Ala228Thr and p.Arg220Gln in PDXK across species. (E) Crystal structure of human pyridoxal kinase with bound adenosine triphosphate (ATP; PDB accession number 3KEU). PDXK is a dimeric enzyme with 1 active site per monomer 65 (monomers A and B are depicted in green and yellow, respectively). In the PDXK structure, the backbone-carbonyl oxygen of alanine 228 establishes a hydrogen bond with the adenine NH 2 group of ATP. The active site of each monomer binds 1 ATP molecule, two Mg 2+ ions, and 1 Na 2+ ion. The ATP-binding site is composed of a β-loop-β structure, often referred as a flap, which provides numerous hydrogen-bond interactions to the ATP βand γ-phosphates, and sequesters the ATP for catalysis. 39 Arginine 220 is located in the β9, in the vicinity of the ATP-binding site. PDB = Protein Data Bank.
August 2019 233 Heterozygote activity (n = 1, F-1-III-1) was within normal range (4.9pmol DBS −1 h −1 ). No correlation of DBS PL kinase activity with age was identified in controls. When segregated according to gender, a difference was identified between control males and females (9 males: mean = 9.2, range = 5.9-14.7; 12 females: mean = 7.1, range = 2.6-11.7) that approached significance (p = 0.1099 by untailed t test). An insertion located in the erythroid-specific promoter region of PL kinase has been associated with reduced erythrocyte PDXK activity. 21 Sequencing this region showed that the reduced activity of all affected individuals was not attributable to this insertion (Supplementary Table 4). 42 Finally, we measured plasma PLP concentrations in all affected individuals. Plasma PLP was greatly reduced in cases carrying p.Arg220Gln and p.Ala228Thr compared to age-matched controls (see Fig 3G, H).
To quantify the axonal damage, we measured the NFL protein in plasma from patients carrying PDXK mutations. NFL, a major axonal cytoskeletal protein, is released into cerebrospinal fluid and blood during axonal breakdown, 43 providing a dynamic biomarker of axonal damage in neurodegenerative disorders, including inherited neuropathies. 44 Both p.Arg220Gln and p.Ala228Thr variants were associated with high NFL (Case F2-II-1, 49.6pg/ml; Case F1-II-5, 25pg/ml), within the range of NFL concentrations published for other axonal CMTs. 44 Supplementation with PLP Is Associated with Biochemical Correction and Clinical Improvement Based on clear pathogenicity of the PDXK mutations leading to low PLP, Case F1-II-5 and F1-II-6 were treated with oral PLP. High doses of pyridoxine are known to be neurotoxic 45 ; therefore, we used PLP, which is hydrolyzed to PL, to increase the upstream PL and force the flux with an enzyme with a high K m based on our results with recombinant p.Ala228Thr protein kinetics. We initiated PLP replacement with 50mg/day PLP, a dose that was commonly used in adults taking PLP as a vitamin supplement with no reported side effects. We aimed to achieve plasma PLP concentration above the normal range by PLP replacement, and this was achieved with 50mg/day. Two weeks after initiating the supplementation, plasma PLP levels had increased considerably without any side effects. PLP levels remained stable at 1, 3, 12, 18, and 24 months on PLP replacement (Fig 4A, B, Table 2), with improvement in symptoms (Supplementary Table 5). Clinical improvement was noted after the first 3 months of PLP replacement in the absence of any additional physiotherapy. Sensory examination and vision remained unchanged in both cases. Main motor improvements were in elbow and wrist flexion/extension and hip and knee flexion/extension, with modest improvement in finger extension (see Supplementary Table 1), leading to 8-and 6-point improvement in the MRC sum score at 18 months on PLP replacement in Case F1-II-5 and Case F1-II-6, respectively. At last examination, 24 months on PLP replacement, patients were ambulating with high-stepping gait, unaided (Supplementary Video). Furthermore, neuropathic pain has completely subsided; neither case requires additional analgesic medication.
These results led to an improvement of 5 points in the NIS score in Case F1-II-5 and 4 points in Case F1-II-6 based on motor, pain, and fatigue subsets. Similarly, the CMT neuropathy score on PLP replacement improved by 4 points and 2 points in Cases F1-II-5 and F1-II-6, respectively (see Supplementary Table 5).
Electrophysiological studies performed 18 months after PLP supplementation have demonstrated no disease progression, with no significant changes in the motor and sensory responses. Whereas the distal motor responses were not significantly different from previous studies, the electromyogram of the proximal muscles showed motor unit recruitment and interference pattern improvement in the proximal muscles (quadriceps and tibialis anterior muscles, biceps and flexor carpi radialis) with activity present in the tibialis anterior that was unrecordable on previous studies.
The clinical response to PLP replacement was further supported by measuring NFL protein in plasma before and after PLP replacement from the PDXK cases in Family 1. A reduction of NFL levels to the normal control range 44 (median = 13pg/ml, range = 11-18pg/ml; see Fig 4C) occurred upon restoration of PLP levels and continued to improve with PLP supplementation. At 24 months with PLP supplementation, the NFL level was 2.5 times lower than before supplementation.

Discussion
Here, we show that biallelic mutations in PDXK lead to peripheral neuropathy with optic atrophy. We have shown previously that several inherited disorders affecting B 6 vitamer metabolism, or resulting in inactivation of PLP, are implicated in neurological conditions. These include PNPO (Mendelian Inheritance in Man [MIM]: 603287), 48 ALDH7A1 (MIM: 266100), 49,50 and PROSC deficiency  51,52 where the lack of PLP in the brain leads to early onset, vitamin B 6 -dependent epilepsy refractory to anticonvulsants. In most cases, delayed vitamin B 6 supplementation results in early death or severe handicap in these disorders. The PDXK activity is tissue-specific under PLPdeficiency conditions. In rats, B 6 deficiency leads to rapid decreases in PDXK activity in the peripheries (liver, muscle, and plasma), with maintained PDXK activity and B 6 supply in the brain. 53,54 This could explain the absence of seizures in humans with PDXK deficiency who present predominantly with peripheral nerve disease and normal CNS function. Furthermore, our PDXK co-expression network analyses from peripheral nerve suggest a link to a module enriched for genes involved in oxidation-reduction processes in which mitochondria play a significant role. PLP is a cofactor for several mitochondrial enzymes, but it is not certain how PLP is transported into mitochondria in humans; in yeast, there is evidence for involvement of Mtm1p. 55 Although we acknowledge that coexpression networks generated from bulk RNA-seq of peripheral nerves are complex to interpret, there is growing evidence for the presence and importance of axonal RNA transport. [56][57][58] Thus, our findings are interesting given the strong phenotypic similarity (optic atrophy and axonal polyneuropathy) to other disorders  608507). Interestingly, within the PDXK regulon from the tibial nerve, we found DHTKD1 already linked to Mendelian disorders and associated with primary peripheral axonal neuropathy. 59,60 This suggests that within this regulon there may be other genes involved in neuronal maintenance that could cause a similar disease phenotype to be discovered. Clinically, the PDXK-related neuropathy presented with mixed sensory and motor involvement in early childhood and was associated with optic atrophy later in adulthood. Electrophysiologically, one of the earliest signs of the disease was the absence of sensory responses even when the motor responses were within normal limits, as seen in Family 2. We note that the patients with longstanding disease (Family 1) had significant standing and walking difficulties despite a mild proximal motor deficit most likely due to the combination of severe distal motor involvement with severe proprioceptive impairment, extending to the proximal joints, exacerbated by the severely impaired vision. 61 Based on our collective experience in rare diseases, we acknowledge that it is difficult to implement randomized case-control studies for these patients and the subsequent limitations when designing therapeutic interventions. We note that some clinical improvements seen in the patients reported here, such as pain and fatigue, are susceptible to placebo effects. Furthermore, the severity of the polyneuropathy in the cases that received PLP supplementation and the absence of more proximal nerve conduction studies are limitations for the characterization of the neurophysiological phenotype and response to treatment. However, although we are limited in our ability to fully explain the mechanism of clinical improvement, the supplementation with PLP was associated with normalization of the biochemical profile and improved clinical outcomes in individuals F1-II-5 and F1-II-6. The clinical improvement was supported by the normalization of PLP levels, stable electrophysiological picture, and NFL normalization after treatment. Furthermore, animal models of PDXK knockdown, B 6 dietary deficiency, and pharmacological models all showed significant decreases in circulating PLP levels, with reversible adverse effect(s) on the peripheral/sensory nerve similar to the human phenotype observed here ( Table 3). The severe sensory involvement recorded at least 20 years before treatment most likely accounted for the absence of a clinical effect on the sensory function in the 2 patients from Family 1. We are limited in our ability to fully quantify the potential for electrophysiological improvement at this stage in the disease due to severe axonal loss, the short interval of PLP replacement, and the absence of more proximal nerve conduction studies. Therefore, larger studies with treatment started earlier in the disease are warranted to assess the full potential benefit from treatment.

Conclusions
We show that biallelic mutations in PDXK cause autosomal recessive axonal peripheral polyneuropathy leading to disease via reduced PDXK enzymatic activity and low PLP. We show that the biochemical profile in affected individuals can be rescued with PLP supplementation and that this is associated with improvement in clinical scales. Therefore, we recommend that PDXK mutations should be screened for in patients with autosomal recessive early onset polyneuropathy, and supplementation with PLP should be started promptly with long-term monitoring of clinical outcomes. Furthermore, B 6 vitamers can be linked to diseases through a wide range of processes and genes involved in the vitamin B 6 pathway. Therefore, our results can be extended to other neuropathies of different etiology characterized by reduction of PLP levels or reduction of PLP-dependent enzyme activities. Collectively, these data expand the genetic causes of primary polyneuropathy and the heterogeneity of inborn errors of vitamin B 6 metabolism and identify PLP as a potential therapeutic target for this disorder.